Accumulation and Difference Features

Medical Image Analysis Project - 20/Feb/2019 - Pankaj Kabra

In accumulation density features

  • First image is the normal Accumulation image
  • Second image is the time weighted Accumulation image

In Cosnecutieve Difference features

  • First image is the image at time t
  • Second image is the the image at time t+1
  • Third image is their difference
In [5]:
import os
from skimage import io, img_as_float
import matplotlib.pyplot as plt
plt.rcParams["figure.figsize"] = (15,5)
%matplotlib inline
from matplotlib.patches import Rectangle
import numpy as np
import utils_ml4mi as utl
In [6]:
def get_img_list(patient):
    yolo_bleeding_path = "yolo_data/"
    images = []
    bboxes = []
    for file in os.listdir(yolo_bleeding_path):
        if (".jpg" in file) and ("_orig" in file) and (file.startswith(patient)):
            img = io.imread(yolo_bleeding_path+file)
            images.append(img)
        if (".txt" in file) and ("_orig" in file) and (file.startswith(patient)):
            with open(yolo_bleeding_path+file,"r") as f: bbx = f.read()
            bbx = [int(float(i)*800) for i in bbx.split()[1:]]
            bboxes.append(bbx)
    return images, bboxes
In [78]:
def yolo_bbx_viz(image, bbx, figs, img_name):
    """
    expects that the bbox is in xymiwh format, size of image is 800x800
    """
    fig1, ax1 = plt.subplots(figsize = figs)
    # Rectangle takes bottom-left coords (bottom in terms of data coordinates, as the y axis is flipped, its top left visually)
    bottom_left = (bbx[0]-bbx[2]/2.,bbx[1]-bbx[3]/2.)
    ax1.add_patch(Rectangle(xy=bottom_left, width=bbx[2], height=bbx[3],linewidth=1,edgecolor='r',facecolor='none'))
    ax1.imshow(image, cmap = "gray")
    ax1.set_title(img_name,fontsize=18)
In [84]:
imglist, bbxlist = get_img_list("60_")

aggimg = np.zeros((800,800),dtype="uint64")
aggimg_wgh = np.zeros((800,800),dtype="uint64")
for i,j in enumerate(imglist):
    j = np.array(j, dtype="uint64")
    aggimg += j
    aggimg_wgh += (i+1)**5*j
In [85]:
yolo_bbx_viz(aggimg, bbxlist[-1], (10,10),"Normal Average")
In [86]:
yolo_bbx_viz(aggimg_wgh, bbxlist[-1], (10,10),"Time Weighted Average")
In [87]:
imglist, bbxlist = get_img_list("75_")
aggimg = np.zeros((800,800),dtype='uint64')
aggimg_wgh = np.zeros((800,800),dtype='uint64')
aggdiv = 0
aggwdiv = 0
for i,j in enumerate(imglist):
    #print(i,aggimg_wgh[151:152,151:152],j[151:152,151:152])
    j = np.array(j, dtype="uint64")
    aggimg += j
    aggimg_wgh += (i+1)**5*j
In [88]:
yolo_bbx_viz(aggimg, bbxlist[-1], (10,10),"Normal Average")
In [89]:
yolo_bbx_viz(aggimg_wgh, bbxlist[-1], (10,10),"Time Weighted Average")
In [90]:
imglist, bbxlist = get_img_list("15_")
aggimg = np.zeros((800,800),dtype='uint64')
aggimg_wgh = np.zeros((800,800),dtype='uint64')
aggdiv = 0
aggwdiv = 0
for i,j in enumerate(imglist):
    #print(i,aggimg_wgh[151:152,151:152],j[151:152,151:152])
    j = np.array(j, dtype="uint64")
    aggimg += j
    aggimg_wgh += (i+1)**5*j
In [91]:
yolo_bbx_viz(aggimg, bbxlist[-1], (10,10),"Normal Average")
In [92]:
yolo_bbx_viz(aggimg_wgh, bbxlist[-1], (10,10),"Time Weighted Average")

Consecutive Difference Features

In [98]:
imglist, bbxlist = get_img_list("44_")
In [99]:
yolo_bbx_viz(imglist[-1], bbxlist[-1], (10,10), "Image at time t")
In [100]:
yolo_bbx_viz(imglist[-2], bbxlist[-2], (10,10), "Image at time t+1")
In [101]:
yolo_bbx_viz(imglist[-1].astype('int64')-imglist[-2].astype('int64'), bbxlist[-1], (10,10), "Difference Image")
In [102]:
imglist, bbxlist = get_img_list("71_")
len(imglist)
Out[102]:
18
In [103]:
yolo_bbx_viz(imglist[-9], bbxlist[-1], (10,10),"Image at time t")
In [104]:
yolo_bbx_viz(imglist[-10], bbxlist[-1], (10,10),"Image at time t+1")
In [105]:
yolo_bbx_viz(imglist[-8].astype('int64')-imglist[-9].astype('int64'), bbxlist[-1], (10,10),"Difference Image")
In [106]:
imglist, bbxlist = get_img_list("35_")
len(imglist)
Out[106]:
11
In [107]:
yolo_bbx_viz(imglist[-5], bbxlist[-1], (10,10), "Image at time t")
In [108]:
yolo_bbx_viz(imglist[-6], bbxlist[-1], (10,10), "Image at time t+1")
In [109]:
yolo_bbx_viz(imglist[-4].astype('int64')-imglist[-5].astype('int64'), bbxlist[-1], (10,10), "Difference Image")